The next UI is not a chatbot. It is shared attention.
The blank chat box is powerful, but it is also a strange place to begin when the question started somewhere else.
Most work questions already have a location. A formula is failing in a sheet. A layer looks wrong in a design file. A confusing setting is sitting in front of you. The problem is not that you cannot explain yourself. The problem is that the helpful context is on the screen, and the interface asks you to carry that context somewhere else first.
Why the cursor matters
The cursor is already how people say "this." It is precise, temporary, and user-controlled. A screen companion that respects that gesture can avoid a lot of the unease around screen tools because the user chooses when the system pays attention.
That is a different posture from a tool that silently watches everything or an agent that takes over the task. The default should be help, not replacement. Guidance, not surrender.
The product promise
openhawk is being built for moments when you are stuck:
- point at the thing on your screen,
- ask what to do next,
- get a clear step in context,
- keep control while you do the work.
This is not a claim that every workflow is solved. It is a direction: useful help should meet the user at the point of confusion, and it should leave when the moment is over.
Manners come first
Screen companions need higher trust standards than ordinary apps. They should be visible when active, off by default, honest about what is sent to a model, and clear about what is remembered.
The first openhawk builds will stay conservative about claims. If a cloud model is used, that should be visible. If memory exists, delete controls should be obvious. If an app should be excluded, the user should be able to exclude it.
What comes next
The first useful proof is small: one real workflow, one visible active state, one clear moment where pointing and asking feels easier than writing a prompt from scratch.
That is the site of the work now.